Energy Efficient Cloud Data Center
Praful Anand
Department of Computer Science and EngineeringNational Institute of Technology Rourkela
Rourkela-769008, Odisha, IndiaMay 2015
Energy Efficient Cloud Data Center
Thesis submitted in partial fulfillment of the requirements for the degree of
Master of Technologyin
Computer Science and Engineering(Specialization: Software Engineering)
by
Praful Anand(Roll No.- 213CS3179)
under the supervision of
Prof. A. K. Turuk
Department of Computer Science and EngineeringNational Institute of Technology Rourkela
Rourkela, Odisha, 769 008, IndiaMay 2015
Department of Computer Science and EngineeringNational Institute of Technology RourkelaRourkela-769 008, Odisha, India.
Certificate
This is to certify that the work in the thesis entitled ” Energy Efficient CloudData Center ” submitted by Praful Anand is a record of an original research work
carried out by him under our supervision and guidance in partial fulfillment of the re-
quirements for the award of the degree of Master of Technology with the specialization
of Software Engineering in the department of Computer Science and Engineering, Na-
tional Institute of Technology, Rourkela. Neither this thesis nor any part of it has been
submitted for any degree or academic award elsewhere.
Prof. Ashok Ku. TurukAssociate ProfessorDepartment of CSE
Place: NIT,Rourkela-769008 National Institute of TechnologyDate: 26-05-2015 Rourkela-769008
Acknowledgment
First of all, I would like to express my deep sense of respect and gratitude towards my
supervisor Prof Ashok Kumar Turuk, who has been the guiding force behind this work. I
want to thank him for introducing me to the field of Cloud Computing and giving me the
opportunity to work under him. His undivided faith in this topic and ability to bring out
the best of analytical and practical skills in people has been invaluable in tough periods.
Without his invaluable advice and assistance it would not have been possible for me to
complete this thesis. I am greatly indebted to him for his constant encouragement and
invaluable advice in every aspect of my academic life. I consider it my good fortune to
have got an opportunity to work with such a wonderful person.
I wish to thank all faculty members and secretarial staff of the CSE Department for
their sympathetic cooperation.
During my studies at N.I.T. Rourkela, I made many friends. I would like to thank
them all, for all the great moments I had with them.
When I look back at my accomplishments in life, I can see a clear trace of my family’s
concerns and devotion everywhere. My dearest mother, whom I owe everything I have
achieved and whatever I have become; my beloved father, for always believing in me and
inspiring me to dream big even at the toughest moments of my life; and my brother and
sister; who were always my silent support during all the hardships of this endeavor and
beyond.
Praful Anand
213CS3179
iii
Abstract
Cloud computing has quickly arrived like a deeply accepted computing model. Still,
the exploration and investigation on cloud computing is at a premature phase. Cloud
computing is facing distinct issues in the field of security, power consumption, software
frameworks, QoS, and standardization. The management of efficient energy is one of
the most challenging research issues. The key and central services of cloud computing
system are the SaaS, PaaS, and IaaS. In this thesis, the model of energy efficient cloud
data center is proposed. Cloud data center is the main part of the IaaS layer of a cloud
computing system. It absorbs a big part of the aggregate energy of a cloud computing
system. Our goal is to supply a better explaining of the design issues of energy manage-
ment of the IaaS layer in the cloud computing system. Servers and processors are the
main component of the data center. Virtualization technologies that are the key features
of the cloud computing environment provide the ability for migration of VMs between
physical servers of the cloud data centre to improve the energy efficiency. This is called
dynamic server consolidation that has direct impact on service response time. Energy
efficient cloud data center reduces the overall energy consumed by the data center. This
results in, reduction of cost incurred by the data center, long life of hardware components,
green IT environment, and making more user friendly. Many VM placement techniques,
server consolidation techniques have been proposed. They do not show optimal solution
in every circumstances. They show optimum result only for a certain data set. They did
not consider both VM placement and its migration simultaneously. They did not attempt
to minimize the VM migrations during server consolidation. Still, forceful consolidation
can result in the performance degradation and may lead the SLA negligence. So, there is
a trade-off between performance and energy. A number of heuristics, protocols and archi-
tectures have explored and investigated for server consolidation using VM migration to
reduce energy consumption. The primary objective is to minimize the overall energy con-
sumption by servers without violating the SLA. Our proposed model and scheme show
the better result at most of the data set. It is based on virtualization technique, VMs,
their placement and their migration. Our study focuses on problems like huge amount
of energy consumption by server and processor. So, here energy consumption is reduced
without violating SLA and to meet certain level of QoS parameters. Server consolidation
is performed with minimum number of VM migration. Here, maximum utilization of re-
iv
sources is tried to achieve, but utilization of resources is not compared with the existing
scheme. Our scheme may show different better result for different configuration of the
data center for the same data set. Problem is formulated as a knapsack problem. Pro-
posed scheme inherits some feature from heuristics approach like BF, FF, BFD, and FFD.
These are used for greedy-bin-packing problem. For simulation, input data set is taken
as random value. These random values are general data set used in real scenario and by
the existing scheme. From simulation, it is found that proposed model is achieving the
desired objectives for a number of data set, and for another data set, some percentage loss
of objectives is occurring.
Keywords: Server consolidation, Cloud data center, Cloud computing, Energy effi-
ciency, IaaS, Resource utilization, VM migration, Green IT, Virtualization.
v
Contents
Certificate ii
Acknowledgment iii
Abstract iv
List of Figures viii
List of Tables ix
1 Introduction 11.1 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Architecture of cloud computing . . . . . . . . . . . . . . . . . . . . 4
1.3 Issues and Challenges . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.1 Organizational Change . . . . . . . . . . . . . . . . . . . . 6
1.3.2 Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.3 Security, Legal, And Privacy . . . . . . . . . . . . . . . . . . 7
1.3.4 Issues in cloud computing to support IoT . . . . . . . . . . . . 8
1.3.5 Other Issues . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.6 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 VM Placement and Online Server Consolidation in the Cloud Data
Center 122.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Research Background . . . . . . . . . . . . . . . . . . . . . . . . . 15
vi
CONTENTS CONTENTS
2.2.1 Random Data Selection . . . . . . . . . . . . . . . . . . . . 15
2.2.2 Heuristic Algorithm . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Proposed Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.1 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.2 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.3 Processor Model . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.4 Server Model . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.5 Proposed Algorithm for Server Consolidation . . . . . . . . . . 19
2.4 Simulation and Result . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.1 Simulation Environment . . . . . . . . . . . . . . . . . . . . 19
2.4.2 Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3 Server Consolidation using VM Migration 243.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Research Background . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2.1 Configuration of Data Center . . . . . . . . . . . . . . . . . . 26
3.3 Proposed Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.1 Problem Statement . . . . . . . . . . . . . . . . . . . . . . 27
3.3.2 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.3 Constraints for VM migration . . . . . . . . . . . . . . . . . 28
3.3.4 Relative Workload for Server and VM . . . . . . . . . . . . . 30
3.3.5 Algorithm for Server consolidation . . . . . . . . . . . . . . . 31
3.3.6 Number of Used Nodes, Migration Efficiency, performance degra-
dation of VM, Utilization and Efficiency . . . . . . . . . . . . 31
3.4 Simulation and Results . . . . . . . . . . . . . . . . . . . . . . . . 33
3.4.1 Simulation Environment . . . . . . . . . . . . . . . . . . . . 33
3.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4 Conclusion and Future Work 39
Bibliography 41
vii
List of Figures
1.1 Detailed Cloud Computing Architecture . . . . . . . . . . . . . . . . 5
1.2 Abstract Cloud Computing Architecture . . . . . . . . . . . . . . . . 5
2.1 Graph between no of processors used and save in energy in different time. 20
2.2 Graph between no of pats powered off and save in energy in different time. 21
2.3 Graphical view of result of Table 2.5 . . . . . . . . . . . . . . . . . . 22
2.4 Graphical view of result of Table 2.6 . . . . . . . . . . . . . . . . . . 23
3.1 Multiple Services assigned on (a) Dedicated Servers with Multiple Sched-
ulers (b) Consolidated Servers with One Hierarchical Scheduler . . . . . 24
3.2 Before Server Consolidation . . . . . . . . . . . . . . . . . . . . . . 25
3.3 After Server Consolidation . . . . . . . . . . . . . . . . . . . . . . 25
3.4 Graph between No of VMs and No of Used Nodes . . . . . . . . . . . 33
3.5 Graph between No of VMs and No of Released Nodes . . . . . . . . . 36
3.6 Graph between No of VMs and No of Migrations . . . . . . . . . . . . 37
3.7 Graph between No of VMs and Efficiency . . . . . . . . . . . . . . . 37
viii
List of Tables
2.1 Each server’s maximum load capacity and energy consumption in unit time 15
2.2 Each server’s maximum load capacity and energy consumption in unit
time for another data set . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 No of processors currently using and corresponding save in energy. . . . 19
2.4 No of parts of a chip is powered off and corresponding save in energy. . . 20
2.5 Result for a chain of requests with configuration of data center as Table 2.1 20
2.6 Result for a chain of requests with configuration of data center as Table 2.2 21
3.1 State of Cluster of Server . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 VM’s Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3 VM’s Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4 Comparison of Results between FFD and NodMig . . . . . . . . . . . 34
3.5 Efficiency of NodMig and performance degradation of VMs at different
value of UTC and mAllowed . . . . . . . . . . . . . . . . . . . . . 35
ix
CHAPTER 1
INTRODUCTION
Today, Cloud computing is an emerging technology in academia and industry. It is
adopted as research, course work, and providing different applications to customers as
services. Cloud computing is a very novel computing technology, a distributed kind of in-
ternet based computing in which the resources like infrastructure, platform, software etc.
are provided by the cloud service providers as a scalable, reliable, fault tolerant service to
the customers on their request on a pay per usage basis and provides advantages like on-
demand access, broad network access, rapid elasticity etc. Cloud computing is inherited
from distributed computing with virtualization as additional technique. With virtualiza-
tion traditional data center is converted into cloud data center. 3 key characteristics of
Cloud computing are virtualization, pay-on-demand, and scalability. It is also called as
utility computing due to payment on incremental and request foundation model. Compa-
nies like Amazon Web Service (AWS), Google, Salesforce.com, IBM, Microsoft and Or-
acle have converted their traditional data center to cloud data center. As a starter Amazon
is the one who started giving cloud services in 2006 with Elastic Compute Cloud(EC2)
Amazon’s total revenues are 61 billion dollar followed by Microsoft and Rackspace.com.
Google has started with Google compute engine, which is much faster, than AWS.
Virtualization enables infinite capacity in the cloud data center. It supports better
isolation between services by encapsulating concurrent services into different VMs and
manageability. Virtualized resources, dynamic deployment of VMs, on-demand resource
provisioning among the hosted VMs, and so on are applying virtualization. These works
result in the improvement of the performance of the virtualization, resource utilization,
and consuming energy. So, Virtualization is being extensively used for server consolida-
tion in the cloud computing.
A company or organization having large infrastructure with virtualization save energy
68-87% and results in reduction of carbon emission. It is an approach towards green en-
1
Chapter 1: Introduction
vironment. Now, Companies in U.S would get the saving in energy of 11.0 billion dollar
and reduction in carbon emission of 83.5 million metric tonnes per year by 2022. It is
comparable to over 14.6 million vehicle’s carbon emission per year. Cloud data center
provides services at a lower cost than traditional data center. Cloud data center has few
applications , homogeneous h/w environment , standardized management tools and s/w
architecture , simple workload , and minimal application updating and patching. Infras-
tructure layer is the most important layer of the cloud computing architecture. Energy The
management of energy or to build energy model is one of the demanding issue in the cloud
data center of the infrastructure layer. For building energy efficient cloud data center, we
will optimize each and every component of the data center e.g. optimize processor, server,
memory, hard disk, and cooling system. So, different techniques are proposed for energy
efficient cloud data center. Server consolidation with VM migration, Optimal VM place-
ment, Energy efficient resource allocation are some of them and these are discussed here.
Energy efficiency is achieved by running minimum number of physical servers within
data center with minimal and acceptable violation of SLA. There are many QoS for cloud
computing. QoS are many for many users with various profiles and different needs. QoS
like throughput, delay, jitter, loss rate, bandwidth etc is highly dependent on application
and image resolution, sound quality, appropriate language etc are dependent on the end
user. There is a trade-off between these QoS.
1.1 Literature Survey
In this section, a review on the study of different techniques for server consolidation us-
ing VM migration is presented. Different authors applied different strategies for server
consolidation and VM placement with some variation in their objectives. They created
their own model for server consolidation. They took their data from different sources like
google data center, TU Berlin university, amazon data center etc. or with mathematical
equation like Poisson Distribution, cumulative inverse function distribution, cumulative
distribution function, t-distribution, normal distribution, Gaussian distribution etc and ap-
plied their model on these data.
2
Chapter 1: Introduction
Author Work DoneYao etal.[31]
They defined DVS model. Jobs are scheduled in finite and specified pe-riod of time on optimal processor.
Cheng andGoddard[6]
They considered input/output resources. This results in larger process-ing, and computing time, additional power consumption on system andprocessor.
Qiu etal.[23]
They suggested an algorithm for parallel loops across n number of pro-cessors. Processors’s voltage and frequency were changed according tothe current workload. Their model minimized efflux, changing, and pro-gressive energy through DVS and ABB. Their model strictly considereddeadline, priority and sequence defined for task operations.
Chen etal.[4][5]
They took account a framework for tasks that have specified constraintson homogeneous multi-processor energy-efficient processor and sug-gested a 1.13-approximation algorithm. They also suggested a 1.283-approximation efflux-awake algorithm for only tasks having only timingconstraint to minimize energy consumption.
Pinheiro etal.[22]
They proposed LC technique. LC solves the problem of overloaded andunderloaded servers by distributing the load among servers dynamically.In such a way underloaded server is released and switched off.
Jejurikar etal.[13]
CPU frequency is increased, CPU voltage is decreased, CPU is shutdownin case of no workload to minimize power consumption.
Zhu etal.[33]
Suggested a scheduling algorithm. Slack is shared among n number ofprocessors of embedded system. A queue that contains ready tasks isdeclared globally. Tasks are selected from queue in such a way thatunutilized time of one task is utilized to increase the completion timeof another task by slowing down the speed of CPU.
Heath etal.[11]
Considered the heterogeneous environment within the cloud data-center.
Elnozahy etal.[8]
power supplies are more efficient relatively low. Their work investigatedand explored IVS and CVS.
Younge etal.[32]
In a multi-core system, they showed that increase in the number of coresdoes not proportionate to the rise in energy consumption.
Rodero etal.[24]
They minimized energy consumption of the HPC system that have virtualenvironment in online manner.
Laszewskiet al.[30]
suggested a energy-aware algorithm that schedules VMs in a cluster ofserver that is based on DVS.
Stoess etal.[28]
framed a new 2-leveled energy management model that takes into ac-count of VM system based on hypervisor.
Kim andBuyya etal.[15]
They took into account QoS requirements of clients e.g. total cost, softreal time constraints of services of applications for energy aware place-ment of VMs in the data center that is based on DVS.
Kusic etal.[16]
Proposed a model for dynamic allocation of resources in the cloud datacenter.
Meisner etal.[18]
Their technique carried out the cloud data center between low and highpower states frequently to conserve energy among server cluster.
Lee andZomaya[17]
They proposed 2 heuristic based power-aware task consolidation algo-rithms to maximize resource utilization. They took into account energyconsumption of server cluster in both the active and idle state.
Jing etal.[14]
They created a green cloud by state-of-the art technique and proposedmodel for energy conservation in the IaaS layer of cloud.
3
Chapter 1: Introduction
Jyothi etal.[25]
survey on different techniques for energy efficient server consolidationusing VM live migration.
Ferreto etal.[9]
Heuristics approaches and LP formulation rank VMs with constant andstable capacity to restrict the VM migration.
Murtazaevand Oh[20]
Sercon algorithm which minimized the overall number of utilized PMs,and the overall number of VM migration.
Song etal.[26]
They modelled the cooperation between requests with different QoS pa-rameters, and existing power capacity between simultaneous applicationservices. The name of model is utility-analytic model that is based on-internet . Requests and capacities are placed in queue. Their time andperiod for scheduling is forecasted.
Singh andHemalatha[21]
A VM Placement Technique known as BASIP to overcome the issue ofdeadlock by using a banker algorithm with Stochastic Integer Program-ming.
Eslam etal.[19]
VM placement technique to minimize the data transfer time consump-tion between VMs and thus helps in optimizing the overall applicationperformance.
Chebiyyamet al.[3]
Motivation, Benefits, and Proposed algorithm for server consolidation.
Clark etal.[7]
Proposed cost of migration as downtime of service and the overall VMmigration time. These should be reduced.
Spieksmaet al.[27]
Their approach gave optimal solution for some data set and proposed analgorithm that globally optimized their problem but, showed exponentialtime complexity at worst case and a heuristic approach.
Toth etal.[2]
Proposed an algorithm that produced the lower bound for the solution of2DVPP problem.
Hyser etal.[12]
VM placement and its live migration.
Gmach etal.[10]
proactive trace-based workload VM placement, migration controllerscheme, policies, efficiency, and performance metrics to address the effi-cient management of virtualized resources.
Verma etal.[29]
Application VM placement policy in the virtualized system while con-sidering the migration cost and power cost.
Bobroff etal.[1]
minimized the quantity of needed physical resources and the rate of SLAnegligence using dynamic server consolidation and migration of VM al-gorithm.
1.2 Architecture of cloud computing
As shown in Fig.1.2, here the most standard and hierarchical architecture of the cloud
computing is discussed. It consists of the three basic layer IaaS, PaaS, and SaaS. The
bottom most layer is the Infrastructure as a Service, next is the Platform as a Service, and
the upper most layer is the Software as a Service.
But in some places we find one additional layer development as a service(DaaS) layer
between SaaS and PaaS. DaaS deals with mainly web based development tools that can be
shared among community. It is not a new layer and is derived from traditionally delivery
4
Chapter 1: Introduction
of development tools.
IaaS, PaaS, and SaaS layers provide infrastructure (CPU, Memory, Disk, Bandwidth
etc.) as a service, platform (Python, Java, .Net etc.) as a service, software (Gmail,
Youtube, Facebook etc.) as a service to customers respectively on subscription based
and pay-as-you-go model via the internet. 4 general advents of Software as a Service are
poly-items, individual item, curve occupancy, and poly-holders. In Fig.1.1, each layer
and its sub layers with examples, its scope, and concerned fields are discussed.
Figure 1.1: Detailed Cloud Computing Architecture
Figure 1.2: Abstract Cloud Computing Architecture
5
Chapter 1: Introduction
1. Front End:- This part is accessed and seen by clients. Fat client, thin client, and
mobile device these interact with cloud data storage via middleware like web browser.
Thin client equipment runs via the internet. If internet slows down, a single point of
failure is created, and the corresponding equipment is considered useless.
2. Back End:- It is a cloud of various computers, servers, data storage system. These
are responsible for online storage and provide resources among multiple customers. These
can be installed in public, private, community, and hybrid cloud. These should provide
agility, flexibility, scalability, multi customer support, security, traffic control, and proto-
col known as middleware.
3. Cloud based delivery:- IaaS, PaaS, DaaS, and SaaS these are discussed above.
4. Network:- Intranet, Internet as discussed below:
The cloud network layer provides:
1. High bandwidth with low latency that allows users to access their application ser-
vices and data without any interruption.
2. Agile network, that provides ICT application services and enables the accessing of
resources. It can shift frequently and effectively between servers.
3. Network Security is important with multi - tenancy i.e. with multi customer envi-
ronment.
1.3 Issues and Challenges
In this section, Challenges that are being faced by IT organization whose applications, ser-
vices are migrated to cloud are discussed. As cloud services have been replaced current
desktop services, the design, configuration, policies, equipment, plan within organizations
are needed to be change, What future directions should be adopted within IT organiza-
tion? and so on. These are requirements for studying this section for running business in
optimum way.
1.3.1 Organizational Change
1. Cloud service providers have no custom supports to the customers. So, organization
needs a central IT department that fully concentrates on providing supports of their ser-
vices and products to customers.
2. Cloud service providers have not high level. Their levels are same as before migration
6
Chapter 1: Introduction
to the cloud. So, they need to change their working processes.
3. System administrators have control over some aspect of cloud based system. So, politi-
cal rules and system administrators should not interfere on the cloud based system. Cloud
service provider will directly or indirectly contact with end users.
1.3.2 Costs
1. Generally, hardware, software, and IT support costs are considered. Economic policies
for application migration, procurement policies play a major role to accommodate the
application of cloud computing at the enterprise level. These should collaborate with
economists and business management schools and colleges to make cloud computing at
the enterprise level.
2. Costs, Operational budgets, and Procurement costs need specific signatories to approve
the deployment of enterprise in the cloud computing. But this leads against on-demand
systems. So, enterprise may adopt cost limits and predict usage patterns, and IT usage
peaks within the organization.
3. Allocation of costs to organization in isolation. This issue can be resolved by auditing,
optimizing financial regulations, and developing tools and techniques.
1.3.3 Security, Legal, And Privacy
Individual manger worry about the failure of delivery of the products that have been
promised. Security, legacy, and privacy issues are the major risks in the cloud computing
environment. So, risk avoidance is needed. Other security issues are data loss, phishing,
multi-customer algorithm, and chain of computing components in cloud computing. They
need novel technique to resolve these issues.
1. Enterprises to be alert about security and regulatory issues of their applications that
had shifted to the cloud. This is due to the complexity of their system, moving their some
data under jurisdiction while some move to the cloud.
2. The model of legal and security issues may be developed in the near future by
our cloud service providers and researchers. These model will be eventually granted and
followed by our government. But, currently DOS attack is critical problem in the cloud
computing. Research is ongoing to resolve the problem of the DOS attack.
3. The system that has been deployed in the cloud has less security due to the lack
of control. Legal issues should be resolved before applying cloud computing technique.
7
Chapter 1: Introduction
Security issues when data is moving. This cab be eliminated by applying constraints.
Time and cost incurred in data migration is a major concern. So, the practical issues that
influence data migration should be investigated.
4. When sensitive data of a sector is moving outside the organization, certain require-
ments and regulations of sector should be followed and should not involved any risks.
5. When the company uses in-house and cloud based systems then there is a need to
control the behavior of every component of the organization. Every user should be aware
by these issues. These challenges can be solved by moving every application and service
of the company on the cloud.
1.3.4 Issues in cloud computing to support IoT
Here, some are challenges that are requirement for IoT.
1. To provide dynamic resources for the system enable application flexibility.
2. Testing for the system ensures QOS parameters to support real time needs.
3. The scalability of IaaS layer enables required exponential growth of requests.
4. Reliable structure of cloud computing provides the availability of applications.
5. Privacy and security issues enhance the data protection and user privacy.
6. The efficient and effective management of energy resources mean efficient power
consumption of applications.
7. The federation in the cloud environment executes the applications close to end
users.
8. Interoperable and portable cloud establishes open cloud ecosystem with concerned
characteristics.
1.3.5 Other Issues
Migrating to the cloud reduced infrastructure cost but increased communication cost.
The flexible chain of pool made cost analysis more complicated. Re-design and re-
development of the model is needed. Adding new features require more security, cost,
more customization. So, viable strategy is needed to SaaS cloud provider for portability
and sustainability. SLA specification should provide maximum percentage of intentions
of customer and be easy to verify, evaluate, and enforced by resource allocation policy.
Different layers e.g. IaaS, PaaS, and SaaS provide different SLA specifications. So, more
advanced SLA specification need customer feedback for customized evaluation of SLA in
8
Chapter 1: Introduction
future. There are a number of services and products of the organization. So, the migration
of which one enables more security of its own. Hazy Cloud makes a problem for users
to select a cloud. Interoperability solves this problem. Many steps of tier are involved
in interoperability. First, develop the computing components and IT resources. Second,
deploy many bordering methods to cloud services. Standardization helps to achieve inter-
operability.
1.4 Motivation
1. Information can be accessed by multiple computing devices. 33% companies adopted
cloud computing. It has 33% mobility and 17% cost effective characteristics.
2. A very few companies went to down, while most of them is growing exponentially
after adopting cloud computing.
3. A large number of business firms that are 82% of all companies are saving money with
cloud. But, saving is not so much.
4. Current business is going towards cloud. 65% of companies have subscribed for cloud
computing last one year.
5. It has reduced energy consumption that has resulted in the reduction of carbon emis-
sion and making a green environment.
6. 93% of companies that adopted cloud computing technique have been grown its one or
more field of IT department.
7. 52% of them noticed their increased efficiency of data center, and utilization of re-
sources. 47% of them observed their low operating cost.
8. 80% of them noted their IT department improvement within six months. 74% of small
scale business there is no resistance to move to cloud within the organization.
9. Many countries are preparing their employees for cloud adoption. In which 97% of
Brazilian companies prepared their employees.
10. Very less change occurred in the data security policies after adopting cloud comput-
ing. only 25% companies announced more anxiety after adopting cloud computing. 47%
of Singapore companies have more worry while 47% of Brazilian companies have less
worry about their data security policies.
11. 50% of USA government IT workers are working towards cloud related fields. 48%
USA government agencies moved one or more work flow to the cloud following ”cloud-
first” policy.
9
Chapter 1: Introduction
12. Cloud data center spends vast amount of energy in a cloud computing system.
13. Amazon’s Data Center consumes 42% cost of the total budget in which 53% costs are
incurred by only servers of data center.
14. More power requirement incurs more cost to set up a cloud system, reduces profit of
cloud service provider, makes environment polluted, heating of the h/w component.
15. To achieve energy efficient data center there should be a good knowledge of cloud
computing, data center, virtualization etc.
16. Around 30% servers are under-utilized but spend remarkable quantity of energy per
year.
17. Virtualization enables infinite capacity in cloud data center. Cloud data center pro-
vides services at a lower cost than traditional data center.
19. Cloud data center has few applications, homogeneous h/w environment, standardized
management tools and s/w architecture, simple workload, and minimal application updat-
ing and patching.
20. Cloud customer faces the problem of protection of their own data and application.
This will largely determine whether, and upon which underlying conditions, business-
related sensitive data and applications may be stored in the cloud. So, different security
measures are taken for private and public cloud depending on their threats profile. In IaaS
layer, There is a need for improved pass of money, base for unknown provision planning,
clear metering, and the administration of self-service.
1.5 Objectives
1. Minimize energy consumption of data center without violating SLA.
2. To meet a certain limit of QoS parameters.
3. Maximum utilization of resources.
4. Maximum number of requests can be serviced.
5. Minimize the number of used PMs.
6. Minimize the number of VMs migration.
7. Minimize the VMs migration time.
8. Maximize the server utilization of cluster.
10
Chapter 1: Introduction
1.6 Thesis Organization
Remaining part of the thesis is constructed as follow:
• In chapter-2 processor and server models are designed to consume minimum energy
without violating SLA or violating SLA with a very less percentage and meet a
certain level of QoS parameters. Processor model is based on DVFS technique,
DVS, and DPM technique. Server model is based on LC technique, virtualization,
dynamic resource allocation, and δ-Advanced-DVS policy. VM placement with
online VM migration algorithm that is based on defined processor, memory, and
server model is proposed to meet the objective. Overall efficiency of the algorithm
is different for different data set.
• In chapter-3 Server consolidation is performed using VM migration. We have some
initial state of the cluster. Model and server consolidation algorithm are proposed
to minimize number of used PMs and number of VMs migration. Nodes are almost
homogeneous and VM’s capacity does not change during server consolidation. En-
ergy is minimized with very less cost incurred by VM migration.
• In chapter-4 overall thesis conclusion and future works are addressed.
11
CHAPTER 2
VM PLACEMENT AND ONLINE SERVER
CONSOLIDATION IN THE CLOUD DATA CENTER
2.1 Introduction
Cloud data center is a major component of the IaaS layer of the cloud computing. It
consumes a huge amount of energy. This incurs to more investment, more infrastructure
and computing cost, less profit to cloud service provider and makes cloud data center not
Eco-friendly. So, a number of research has been done and currently, researches are on-
going to minimize energy consumption by data center. Data center consists of physical
servers, processors, memory, storage, and cooling system. So, optimize each and every
component to reduce energy consumption without violating SLA or violating the SLA to
a threshold value. It is also necessary to keep in mind about QoS parameters like through-
put, bandwidth, delay, jitter, loss rate. The QoS parameters are image resolution, sound
quality, appropriate language etc. for end users. There is a huge number of underutilized
servers in such a data center and create a big issue for cloud service providers. When
dealer’s application services execute in separation to meet security issues. Then it gen-
erally results in under utilization of servers. Physical servers consume a huge amount of
energy for servicing requests. In this chapter a model is proposed to minimize energy con-
sumption by data center with maximum utilization of resources. This model adopts some
properties of heuristic algorithms like FF and BF. FF and BF have objective to minimize
the number of PSs, but our algorithm has objective to maximize the resource utilization as
well. We optimized each component by applying existing and our proposed techniques.
We proposed a algorithm for server consolidation using VM migration technique. Server
consolidation means that service the requests with minimum number of used physical
servers at a time. Server consolidation provides advantages like total reduction in CPU
cycles , total reduction in administration and energy prices, resultant reduction in shifting,
12
Chapter 2 : VM Placement and Online Server Consolidation in the Cloud Data Center
connection, hard disk, memory and repository and reduction in carbon emission from data
center.
Virtualization enables server consolidation in a very efficient and effective way. One
of the major application of virtualization technique is server consolidation. It also sup-
ports maximum utilization of servers. It offers better isolation, manageability, resource
provision on demand basis for server consolidation. Virtualization is a technique that
makes virtual version of anything e.g. OS, Physical servers, Storage, and network re-
sources in IT sector. Resource virtualization, dynamic deployment of VM, resource al-
lotment according to pay-as-you-go model among the assigned VMs, and so on advance
to upgrade in the achievement of resource utilization and virtualization. Performance un-
predictability, proof of requests arrival distribution, and the performance of application
services are phenomenons in virtualization. A similar aspect of a chain of resources is
supported by virtualization. A server of cloud data center can accommodate a number
of computing resources encapsulated as a VM. These VMs are isolated from each other.
It makes applications disjoint from the system’s hardware, and provides simple adminis-
tration of VMs to data center managers. Applications enclosed as VMs can be paused,
terminated, copied, blocked, and provide a virtualization layer between the OS layer and
system’s hardware layer. The state of VM is saved and data package is stored in reposi-
tory as a substitute to repair from catastrophic failure or error. Virtualization is one of the
major feature of cloud computing. Virtualization supports scalability, easy manageability
features of the cloud computing. Here, the hosting of VMs is dynamically exchanged
without preventing their running of services. Virtualization extracts the descriptions of
system’s hardware and supplies virtualized resources for effective and powerful applica-
tions. The virtualization layer like a software is called a VM monitor(VMM) or hypervi-
sor. It enables the physical resources of the server as virtual resources i.e. creates multiple
VMs on the server. Each VM runs for processing of application. Different VMs may have
different computing environment on the same server. VM migration that is a part of the
virtualization enables the load balancing among servers of the cloud data center. VM
live migration that enables shifting of VMs from one server to another server without
preventing the services executing in VM is a key feature of virtualization and provides
benefits like fault tolerance, maximum resource utilization, the overall balancing of cur-
rent workload, and online system maintenance etc. Online VM placement is continuously
optimizing due to changeable workloads supplied to applications. The main causes for
huge energy consumption are huge amount of computing resources, energy inefficient
13
Chapter 2 : VM Placement and Online Server Consolidation in the Cloud Data Center
hardware, and inefficient management of resources. Absolutely idle servers waste about
65% of their crest power. When computing resources consume 1 W of power, an extra
0.3-0.8 W is needed for the cooling system.
Server consolidation improves resource utilization and minimizes energy consump-
tion. Simultaneous application services are encapsulated into different VMs. Thus, these
services are isolated from one another. The variations and inconsistencies on resource
requirements of the simultaneous services provide occasions to server consolidation for
maximize the resource utilization and minimize the energy consumption. We can com-
pare between dedicated and consolidated servers. The current workload is measured and
its achievements are measured in terms of probable value. The workloads are consol-
idated in various ways. Each way results in the saving of different number of servers
and different amount of power. Results of simultaneous application services that have
been affected by virtualization are varying. So, there is an another challenging task to
create a model that definitely realizes the potential revenue of cloud data center. Dynam-
ically turning on/off servers, utility management, and on-demand resource management
are other techniques for server consolidation in the cloud data center to save energy and
progress the QoS parameters. These techniques are active and take decisions during the
execution of application services. The dynamic allocation of resources and the mapping
of VMs are not only techniques to explain the vast usage of virtualization. But, only these
techniques are able to target on the arrangement of scope of the data center that are based
on internet when many application services are to be consolidated within the cloud data
center. So, in this chapter such type of problem and its solution is discussed.
We are again discussing about the relation between resources and application services.
This relation is affected by virtualization. This chapter models a utility analytic model for
server consolidation that is based on internet, defining the relationship between different
requests of service arrival with different QoS parameters, and the potential passing across
simultaneous application services in the cloud data center to solve the above issues. This
model uses a queue and predicts the time for assigning the CPU or resources to the task
and the completion time for task execution. Utility analytic model gives the topmost
constrained of the number of severs after consolidation to grant the QoS parameters but
having equal probability of loss of requests like in the dedicated servers. Simultaneously,
function of power and utilization of PSs, the aerial of virtualization, and the algorithm
for the dynamic allocation of resources in the cloud data center is formulated for the
evaluation of server consolidation.
14
Chapter 2 : VM Placement and Online Server Consolidation in the Cloud Data Center
2.2 Research Background
This section describes the data set used for evaluation of proposed model. Another data
set is used for processor model within the data center. Another thing is that proposed
algorithm is inherited from heuristic algorithms NF, FF, FFD, BF, and BFD.
2.2.1 Random Data Selection
Random data set is normalized to obtain better accuracy with proposed algorithm.
1. We took ten servers with maximum load capacity and energy consumed by them as
shown in Table 2.1.
Table 2.1: Each server’s maximum load capacity and energy consumption in unit time
No of Server S0 S1 S2 S3 S4 S5 S6 S7 S8 S9
Max. load capacity 10 20 30 40 50 60 70 80 90 100Energy Consumed 50 60 70 80 90 100 110 120 130 140
2. We took another ten servers with maximum load capacity and energy consumed by
them as shown in Table 2.2.
Table 2.2: Each server’s maximum load capacity and energy consumption in unit time foranother data set
No of Server S0 S1 S2 S3 S4 S5 S6 S7 S8 S9
Max. load capacity 25 39 79 89 100 110 129 138 145 156Energy Consumed 110 700 770 780 900 950 1000 1005 1010 1020
2.2.2 Heuristic Algorithm
Here, We inherited the features of Best-Fit algorithm, First-Fit algorithm from bin-packing
problem. We sorted servers in increasing load capacity. At the time of migration, we tra-
verse servers from highest load capacity to lowest and put all request to first server that
service requests.
15
Chapter 2 : VM Placement and Online Server Consolidation in the Cloud Data Center
Algorithm 1: Best-FitInput: lV M, lPMOutput: MapVMPM1. foreach VMi ∈ lV M2. AssignIx←−03. foreach PMj ∈ lPM4. if(VMi fits in PMj)5. then rCap[AssignIx]←−Compute RemainCap(VMi, PMj)6. AssignIx←−AssignIx+17. end if8. end foreach9. MinCapIx←−Find IndexMinCap(rCap)10. MapVMPM[i]←−Perform MappingVMPM(rCap, VMi, lPM , MinCapIx)11. end foreach
2.3 Proposed Model
2.3.1 Objectives
1. Minimize energy consumption of data center without violating SLA.
2. To meet a certain limit of QoS parameters.
3. Maximum utilization of resources.
4. Maximum number of requests can be serviced.
2.3.2 Assumptions
1. Each server has different maximum load capacity.
2. The server that has larger load capacity , consumes greater energy independently the
current load servicing by that server.
3. We are not aware of the behavior of the incoming loads.
4. Initially server cluster is in the idle state.
2.3.3 Processor Model
1. Switching off components of the chips by DPM technique.
2. Minimum number of processors should , be run , have excellent performance , ca-
pability per W, more effective workload management , apply the virtualization , and the
potential of running in higher temperature environments.
3. Processors have energy efficient task , job , thread scheduling mechanism.
4. Execute task at critical speed.
16
Chapter 2 : VM Placement and Online Server Consolidation in the Cloud Data Center
5. An optimal loop scheduling and voltage assignment algorithm minimizes both leakage
and dynamic energy using DVS and ABB.
6. Multi-core and Multi-processor system.
7. Slowing down the computing speed at the low workload.
8. Processors have survivability and fault tolerant features.
9. Small and simple 1st level cache and way prediction decrease power consumption.
10. Compiler optimization i.e. any improvement at compile time improves power con-
sumption.
11. No traffic within processor.
Mathematically, We can write one of the equation as follows:
Let one chip has n parts. Each consumes energy as follows:
E1, E2, E3, ....., En
in unit time.
Total Energy = TE =∑n
i=1Ei
Let first k parts are powered off due to currently not using.
Total Energy after powering off k parts = MTE =∑n
i=1+k Ei
So, save in energy = SE = TE - MTE (1)
Another equation is as follows:
This equation is for finding save in energy when we use multiprocessor system. Let, k be
the number of processors in the multiprocessor system.
Energy consumed for executing one task by server = Energy consumed by processor to
execute one task + Energy consumed by other parts of server.
TE = E + RE
Let, We have n tasks.
Energy consumed in executing n tasks by server = n * TE
NTE = nE + nRE.
Let 1 < k < n
Energy for processing of k tasks:
KTE = kE + RE.
Energy for processing of remaining n-k tasks:
RTE = (n-k)E + RE
Total energy consumed for processing n tasks:
17
Chapter 2 : VM Placement and Online Server Consolidation in the Cloud Data Center
ONTE = KTE + RTE
So, save in energy:
SE = NTE - ONTE
= NTE - KTE - RTE
= nE + nRE - kE - RE - nE + kE - RE
SE = (n-2)RE (2)
2.3.4 Server Model
1. Apply LC technique(only improve overall energy consumption , under heavy load does
not work properly).
2. Apply the combination DVS , CVS and cluster reconfiguration technique.
3. Adopt Virtualization and VM live migration across physical servers for energy effi-
ciency.
4. Sever consolidation with VM live migration.
5. Dynamic allocation of resources and DVS.
6. Energy-aware allocation of VMs to provide services in the cloud data center that sup-
port DVS.
7. δ-Advanced-DVS policy to schedule VMs, to reduce energy consumption, to maxi-
mize the acceptance rate of provisioning requests in the real time.
8. Multi-core environment.
9. To drop the power of part of the larger host system that are not needed by the VMs
placed on it.
10. For specific allotment, Optimal resource allotment strategies for specific in the dy-
namic algorithm.
11. Operation overhead is considered in decision making.
12. The explanation and answer must be scalable as the size of cloud data center contin-
uously increases.
18
Chapter 2 : VM Placement and Online Server Consolidation in the Cloud Data Center
Table 2.3: No of processors currently using and corresponding save in energy.
No of proces-sors using
3 4 5 6 7 8 9 10 11
save in energy(kj)
1 2 3 4 5 6 7 8 9
2.3.5 Proposed Algorithm for Server Consolidation
Algorithm 2 Server Consolidation
Step 1: Input: newrequest
Step 2: Assign the request to first server that can service the request. Highlight that
server as maximum utilized or not.
Step 3: Input: newrequest.
Step 4: Assign the request to the first server that is not maximum utilized and can
service the request. If service could not be serviced by any server. Display output as
request can not be serviced.
Step 5: After assignment in Step 4 if server is maximum utilized. highlight that
server. goto Step 3
Step 6: Migrate all VMs that are servicing request to the maximum capacity server
that can service all request.
Step 7: Highlight that server as maximum utilized or not. goto Step 3
2.4 Simulation and Result
In this section, simulation environment, Result obtained from implementation of two equations
and proposed algorithm, and observation from these results are described.
2.4.1 Simulation Environment
We implemented the above two equations using Matlab, ×86 architecture system, Windows 7 as
OS, Intel Core2 processor, 4GB RAM, Windows network elements. Proposed algorithm is imple-
mented in C++. We took data from obtained result and draw the result in graphical view using
matlab with the system having features described above.
2.4.2 Result
Figure 2.2 depicts equation number 1. From Figure 2.2, The observation is as Table 2.4. So, As
the no of parts are powered off due to currently not using save in the energy increases linearly.
19
Chapter 2 : VM Placement and Online Server Consolidation in the Cloud Data Center
Figure 2.1: Graph between no of processors used and save in energy in different time.
Table 2.4: No of parts of a chip is powered off and corresponding save in energy.
No of partspowered off
0 2 4 6 8 10 12 14 16
save in energy(kj)
0 2 4 6 8 10 12 14 16
Figure 2.1 depicts equation number 2. From Figure 2.1, The observation is as Table 2.3. So,
As the no of processors are increasing save in the energy increases linearly.
In Table 2.5 we showed the energy consumption by the data center at every new load for Table
2.1.
Table 2.5: Result for a chain of requests with configuration of data center as Table 2.1
Time T0 T1 T2 T3 T4 T5 T6 T7 T8 T9New Load 34 49 56 12 69 78 23 90 10 15EnergyConsumed
80 78 140 270 270 240 340 340 390 380
In Table 2.6 we showed the energy consumption by the data center at every new load for Table
2.2.
From Figure 2.3 and 2.4 we can see that on increasing load energy consumption by data center
20
Chapter 2 : VM Placement and Online Server Consolidation in the Cloud Data Center
Figure 2.2: Graph between no of pats powered off and save in energy in different time.
Table 2.6: Result for a chain of requests with configuration of data center as Table 2.2
Time T0 T1 T2 T3 T4 T5 T6 T7 T8 T9New Load 20 25 111 100 80 90 40 10 120 28EnergyConsumed
110 1020 1020 1920 2930 3935 3935 3935 4935 4935
does not increase significantly and further observation is as follows:
Energy consumption, overall utilization of the server cluster at any time t depends on the
current load and some set of previous load that has been processed or processing. Max. load
capacity and energy consumed by each server.
Ui = Tu/Tt
Tu = time for which server is used.
Tt = time for which server could be used.
Ei = Eu/Et
. Eu=Energy consumed by one server.
21
Chapter 2 : VM Placement and Online Server Consolidation in the Cloud Data Center
Figure 2.3: Graphical view of result of Table 2.5
Et=Total energy consumed by all server.
Oi = Ui + Ei
2.5 Summary
We reviewed the existing energy efficient techniques for creating the cloud data center. Then we
modeled, energy efficient cloud data center at the IaaS layer of cloud computing, minimized the
energy consumption to a convinced and positive limit and also assisted to decrease the CO2 release
from the cloud data center with considering my objective using some existing techniques, hybrid
techniques, modified techniques and proposed techniques.
22
Chapter 2 : VM Placement and Online Server Consolidation in the Cloud Data Center
Figure 2.4: Graphical view of result of Table 2.6
23
CHAPTER 3
SERVER CONSOLIDATION USING VM MIGRATION
3.1 Introduction
Cloud computing has 3 important properties virtualization, pay on-demand, and scalability. Pay-
on-Demand and scalability provide cloud users to change their resource demand in a very small
time and pay to cloud provider accordingly. The work that we did come in IaaS type. One of
the great application of virtualization is server consolidation in the cloud data center. Virtualiza-
tion enables resources as a chain of resources rather than dedicated for the individual application.
server can run many application enclosed as a VMs with multiple computing environments. In
the result, it provides better management, on-demand resource provision, and isolation. Server
Figure 3.1: Multiple Services assigned on (a) Dedicated PSs with Multiple Schedulers (b)Consolidated PSs with One Hierarchical Scheduler
consolidation is the technique by which VMs are consolidated or assigned to minimum number
24
Chapter 3 : Server Consolidation using VM Migration
of servers within the data center. This is not a type of batch job. VMs are dynamically hosted on
servers according to need and freed periodically. The main motto of Server consolidation is to min-
imize the power consumption by powering off idle PMs within the data center. This results in the
green IT or green computing and reduces the cost in servicing of requests. In Figure 3.2 and 3.3,
server consolidation using VM live migration is clearly shown. In the perspective of scalability,
Figure 3.2: Before Server Consolidation
Figure 3.3: After Server Consolidation
server consolidation increases resource utilization, reduces the problem of under utilization and
over utilization of servers, reduces CPU count, this is not available in the dedicated server where
a server can not be shared by multiple services. In Figure 3.1.a A scheduler assigns one service to
one server, another scheduler assigns a single service to another server, and so on. This ensures
isolation between services, and the wastage of resources and power among cluster of server. In
Figure 3.1.b There is one virtualization layer between scheduler and cluster of servers. Services
are encapsulated as VMs. These VMs are isolated to one another and assigned to servers with
the help of virtualization layer. This ensures maximum resource utilization, minimum power con-
sumption with acceptable degradation in another QoS parameters due to the variation in resource
requirements with time among these VMs. This is done with some loss probability of requests. In
server consolidation, minimum number of VM migration is one of the challenging issue, because
a number of VM migration takes place in achieving minimum number of used nodes. VM live
migration is a costly operation. It consumes some CPU cycles, memory, and network bandwidth.
Some prior works did not consider a large number of VM migrations during server consolidation.
In this chapter, An algorithm named NodMig is proposed. server consolidation is done with mini-
mum number of VM migrations. We took some idea from greedy bin-packing algorithms like BF,
25
Chapter 3 : Server Consolidation using VM Migration
FF, FFD, and BFD. The server that has least workload and contains at least one VM among all
server is selected as a donor server. Rectangle and vector packing are two types of packing prob-
lem. Here vector packing problem is discussed with two dimensions as CPU and memory. Use
of energy efficient hardware, energy minimization in network, server consolidation, energy-aware
scheduling, these are another ways to reduce power consumption within the data center. There are
two types of server consolidation technique offline and online. Online technique is discussed in
chapter 2. Here, offline technique is discussed where we have some intermediary state of cluster of
server to achieve required final state. In online technique, we are not aware the type of upcoming
request or application. while in offline we are aware. Score of nodes is calculated by mathematical
equation in each iteration to find donor node. On donor node, the score of each VM is calculated
by mathematical expression. Two constraints are considered for VM migration between servers.
Metrics are defined to calculate the efficiency of NodMig algorithm. Finally, we implemented the
algorithm in C++ programming language. We took the data from obtained result and presented
graphical results using MATLAB. In implementation, we took a standard data as input. We com-
pared NodMig algorithm with FFD algorithm. Finally, Results of simulation is analyzed and its
effectiveness is calculated with different parameters. Cloud data centers that are based on internet
do not certainly perceive their possible earnings using virtualization for server consolidation.
Note:-Node, PS, PM, Host and Server words are used interchangeably.
3.2 Research Background
This sub-section presents the data set being used for the evaluation of NodMig. NodMig has some
properties of greedy-bin-packing algorithm FFD, BFD, FF, BF. These algorithms are described
here. We took standard data for the configuration of data center. We simulated with the variation
of CPU threshold value and maximum allowed VM migration during server consolidation. Then,
we found the optimal threshold value for CPU utilization and maximum allowed VM migration
during overall server consolidation.
3.2.1 Configuration of Data Center
We took a random normalized data for current state of data center. There are 6 servers and 15
VMs as shown in Figure.3.2 or 3.3. Each server has different CPU and memory capacity.
Table 3.1: State of Cluster of Server
Servers S0 S1 S2 S3 S4 S5
No of VMs 3 3 1 2 2 4VMs V0, V1, V2 V3, V4, V5 V6 V7, V8 V9, V10 V11, V12, V13, V14
26
Chapter 3 : Server Consolidation using VM Migration
Table 3.2: VM’s Capacity
VMs V0 V1 V2 V3 V4 V5 V6 V7CPU Capacity 1.0 1.0 1.5 1.0 1.2 1.0 0.5 0.5Memory Capac-ity
0.5 0.25 1.0 0.25 0.5 1.0 0.5 0.5
Table 3.3: VM’s Capacity
VMs V8 V9 V10 V11 V12 V13 V14CPU Capac-ity
1.0 1.0 1.0 1.5 1.0 1.0 0.5
Memory Ca-pacity
1.0 0.5 0.75 1.0 0.75 0.5 0.25
Algorithm 3 First-Fit-Decreasing Algorithm
Input: V, S
Output: totalMig, totalRelNe, totalUseNe.
1: PointVM[ ]←− VMPoint Calculation(V)
2: SortedVM[ ]←− SortVM byPointDecrg(PointVM, V)
3: foreach VM∈ SortedVM[ ]
4: for j←−0 to |S|-1
5: Result←−CheckForMigration(VM, j)
6: if(Result)
7: then Migrate VM to j
8: break for loop
9: end for
10: end foreach
Where
V=Set of Virtual Machine.
S=Set of Server.|S|=No of Server within Data Center.
j=Index for Server.
3.3 Proposed Model
3.3.1 Problem Statement
N1, N2, N3, N4, ...., andNnN are nN number of physical servers within the cloud data center.
Number of VMs for Nj PM = pj = |Nj |, 1 ≤ j ≤ nN .
27
Chapter 3 : Server Consolidation using VM Migration
Let, the value of bNSj indicates the status of Nj PM.
bNSj =
1 if Nj is an active PM.
0 otherwise.
Min{nN∑j=1
bNSj} (3.1)
R(Nj) indicates resource utilization for Nj .
Max{∑nN
j=1 R(Nj)×bNSj∑nNj=1 bNSj
} (3.2)
Let, the value of bV Mksd indicates the migration status of the kth VM.
Total number of VMs= tnVM =
nN∑j=1
pj∑i=1
1.
bV Mksd =
1 if s6=d.
0 otherwise.
Min{tnVM∑k=1
bV Mksd} (3.3)
3.3.2 Assumptions
1. Multidimensional PMs and VMs.
2. Vector packing problem not Rectangle packing problem.
3. Cluster of servers is not in ideal state. VMs are already placed on servers.
4. Offine algorithm i.e. We are aware of characteristics of upcoming requests.
5. Heterogeneous cloud data center.
6. VM Live Migration.
7. The maximum resource capacity of VMs does not change during server consolidation.
3.3.3 Constraints for VM migration
The ratio of sum of CPU capacity of all VMs on individual server to the maximum CPU capacity of
server must be less than or equal to the threshold value of CPU utilization. Formal representation
is as follows:p∑
i=1
V Cij
Cj≤ UTC , ∀ j (3.4)
28
Chapter 3 : Server Consolidation using VM Migration
where
UTC=Threshold value for utilization of CPU on jth server.
p=Number of VMs on jth server.
V Cij =CPU capacity of ith VM on jth PS.
Cj =CPU capacity of jth PS.
j=jth server.
The sum of capacity of each resource of VMs on server must be less than the capacity of server
for that particular resource. In this model, we are considering 2D type of VM and server. Each
server consists of memory and CPU and each VM is encapsulated as memory and CPU. Formal
representation is as follows:p∑
i=1
V Cij < NCj , ∀ j (3.5)
p∑i=1
VMij < NMj , ∀ j (3.6)
If we consider server and VM that has n resources as n-dimensional vector. Formal condition
is as follows:p∑
i=1
Vij < Nj , ∀ j (3.7)
Eqn.(3.7) is the combination of eqn.(3.5) and eqn.(3.6).
where
j=jth Server.
p=No of VMs on jth server.
V Cij =CPU capacity of ith VM on jth server.
VMij =Memory capacity of ith VM on jth PS.
NCj =CPU capacity of jth PS.
NMj =Memory capacity of jth PS.
Nj =< rN1j , rN2j , rN3j , ......., rNnj >.
Vij =< rV1ij , rV2ij , rV3ij , ......, rVnij >.
If we schedule a VM ViS for live migration from NS to ND. Then server NS and ND should
satisfy the eqn.(3.4) and eqn.(3.7) after the VM live migration. VM ViS is converted to ViD.
Another thing is that NodMig algorithm supports a maximum number of VM migration through
overall server consolidation. So, at each iteration total number of VM migration from the first it-
eration to the current iteration is checked against the maximum number of allowed VM migration.
If total number of VM migration exceeds the maximum number of allowed VM migration then
no more server consolidation takes place. Maximum number of allowed VM migration is directly
proportional to the total number of VMs within the cloud data center. We calculate the maximum
number of allowed VM migration as follows:
29
Chapter 3 : Server Consolidation using VM Migration
mAllowed ∝nN∑j=1
pj∑i=1
1 (3.8)
where
mAllowed =Maximum number of allowed VM migration through server consolidation.
pj=Number of VMs on jth server.nN∑j=1
pj∑i=1
1 = tnVM =Total Number of VMs.
3.3.4 Relative Workload for Server and VM
We are calculating the relative workload of node and its assigned VMs as points. Node’s points
help in finding the source PS and destination PS for VM migration. VM’s points ensure minimum
performance degradation, minimum service downtime, and maximum utilization of server. First,
point of server is calculated. Server’s points are ordered in decreasing order. Least pointed node
that has at least one VM is selected as a source or a donor node. On donor node, VMs are listed
in decreasing order of their points. VMs are migrated from the most pointed to least pointed VM
of the sorted VMs list. VM Migration is performed using eqn.(3.4) and eqn.(3.7). Mathematical
Equation to calculate point for server is as follows:
RWLCj =
p∑i=1
V Cij
Cj, ∀ j (3.9)
RWLMj =
p∑i=1
VMij
Mj, ∀ j (3.10)
where
Mj =Memory capacity of jth server.
RWLCj =Relative CPU workload on jth server.
RWLMj =Relative memory workload on jth server.
The point of jth server is calculated as follows:
pNj =RWLC2
j+RWLM2j
nN∑j=1
RWLC2j +RWLM2
j
, ∀ j (3.11)
where
pNj =Point for jth node.
nN=Number of nodes.
Mathematical Equation to calculate point for VM is as follows:
pVij =V C2
ij+VM2ij
p∑i=1
V C2ij + VM2
ij
, ∀ j (3.12)
where
pVij =Point for ith VM on jth server.
30
Chapter 3 : Server Consolidation using VM Migration
3.3.5 Algorithm for Server consolidation
The following algorithm is for server consolidation using VM migration. The modeling of all
functions that are addressed in NodMig are modeled above in mathematical equations.
Algorithm 4 NodMig AlgorithmInput: mAllowed, N, cVM, cNOutput: totalMig, totalRelNe1. totalMig←−02. totalRelNe←−02. while (mAllowed ≥ totalMig)3. for k←−1 to |N |4. pointN[k]←−compute PointN(N, cN, cVM)5. end for6. pointDecrN[ ]←−calculate pointDecrN(pointN)7. rNI←−releaseNodeIndex(pointDecrN)8. listVMs[ ]←−obtainListVMs(N, pointN, pointDecrN, rNI)9. pointVMs[ ]←−compute PointVM(listVMs)10. pointDecrVMs[ ]←−calculate pointDecrVMs(pointVMs, listVMs)11. nMig←−012. foreach VM ∈ listVMs[ ]13. for k←− 1 to rNI-114. migPos←− Find MigPos(VM, cN, cVM, N, pointDecrN)15. if(migPos)16. then nMig←−nMig+117. scheduleVMMigration[ ]←−create VMMigSchedule(VM, N)18. break for loop19. end for20. end foreach21. if(nMig = |listVMs[ ]|)22. then totalMig←−totalMig+nMig23. totalRelNe←−totalRelNe+123. N←−updatebyVMMig(N, scheduleVMMigration)24. else break(while loop)25. end while
3.3.6 Number of Used Nodes, Migration Efficiency, performance degra-
dation of VM, Utilization and Efficiency
From NodMig algorithm, we find the number of released PSs and the number of total VM mi-
gration after server consolidation. Number of used nodes is equal to the total number of physical
servers minus number of released nodes. These are performance metrics for comparing results
between FFD and NodMig.
NUNi = TNPSi −NRNi (3.13)
Where
31
Chapter 3 : Server Consolidation using VM Migration
TNPSi=Total number of physical servers for ith simulation.
NRNi=Number of released nodes for ith simulation.
NUNi=Number of used nodes for ith simulation.
Migration Efficiency is the efficiency of algorithm in terms of number of VM migration. Less
number and more number of VM migration ensure more and less efficiency of the algorithm with
respect to the VM migration. Migration efficiency is equal to the ratio of difference between the
total number of VMs and number of VM migration to the total number of VMs.
effMigi =tnVMi−tnVMmigi
tnVMi× 100 (3.14)
where
effMigi =Migration Efficiency at ith simulation.
tnVMmigi =Total number of VM migration after server consolidation at ith simulation.
Efficiency of algorithm is calculated in terms of overall number of released PMs and overall num-
ber of VM migration after server consolidation. Overall efficiency is equal to the ratio of number
of released nodes to the number of VM migration.
Effi =NRNi
tnVMmigi× 100 (3.15)
where
Effi =Efficiency of algorithm at ith simulation.
We are calculating degradation in the performance of VM for finding the optimal value of UTC
and the maximum number of allowed VM migration. Degradation in the performance of VM is
the ratio multiple of UTC and migration time for VM to the total execution time for NodMig.
pDegVM =TV Mig
Ttet× UTC × 100 (3.16)
where
pDegVM =Performance degradation in VM.
TVMig =Time taken for VM migration.
Ttet =Execution time for NodMig.
Utilization of individual server and overall active server are calculated as follows:
R(Nj) =∑pj
i=1 Vij
Cj(3.17)
The overall utilization of the cloud data center is as follows:
oUTN =∑nN
j=1 R(Nj)×bNSj∑nNj=1 bNSj
(3.18)
32
Chapter 3 : Server Consolidation using VM Migration
3.4 Simulation and Results
In this section, Simulation Environment, Result obtained from implementation of proposed algo-
rithm, Results in graphical view, and observations from these results are described.
Figure 3.4: Graph between No of VMs and No of Used Nodes
3.4.1 Simulation Environment
We implemented NodMig and FFD in C++, with×86 architecture system, Windows 7 as OS, Intel
Core 2 Duo at 3 GHz processor, 4GB RAM, and Windows network elements. From Table 3.5, we
observe that
mAllowed = (0.75)× tnVMi (3.19)
UTC = 0.85
Result is shown in Table 3.4. We took the data from Table 3.4 and draw the result in graphical
view as shown in Fig. 3.4, Fig. 3.5, Fig. 3.6, and Fig. 3.7 using MATLAB with the system having
features described above.
33
Chapter 3 : Server Consolidation using VM Migration
Table 3.4: Comparison of Results between FFD and NodMig
FFD NodMigNo ofVMs
No ofUsedNodes
No ofRe-leasedNodes
No ofMigra-tions
Efficiency No ofUsedNodes
No ofRe-leasedNodes
No ofMigra-tions
Efficiency
25 11 4 22 18.18 10 5 17 29.4150 21 10 47 21.28 20 11 36 30.5575 31 14 72 19.44 30 15 47 31.90100 41 20 97 20.62 40 21 65 32.30125 51 24 122 19.67 50 25 76 32.90150 61 29 147 19.73 60 30 90 33.33175 71 33 172 19.19 70 34 100 34.00200 81 39 197 19.80 80 40 116 34.48225 91 43 222 19.37 89 45 129 34.88250 101 49 247 19.84 99 51 145 35.20275 111 53 272 19.48 109 55 154 35.71300 121 59 297 19.86 119 61 169 36.09325 131 63 323 19.50 129 65 177 36.72350 141 68 348 19.54 139 70 188 37.23375 151 72 373 19.30 149 74 195 37.95400 161 78 398 19.60 159 80 208 38.46425 171 82 423 19.38 168 85 217 39.17450 181 86 448 19.20 178 89 223 39.91475 191 92 473 19.45 188 95 235 40.40500 201 96 498 19.28 198 99 241 41.08525 211 102 523 19.50 208 105 251 41.83550 221 106 548 19.34 218 109 258 42.25575 231 111 573 19.37 228 114 266 42.86600 241 115 598 19.23 238 118 273 43.22625 251 117 623 18.78 248 120 273 43.96650 261 119 648 18.36 258 122 274 44.52675 271 121 673 17.98 268 124 276 44.93700 281 123 699 17.60 277 127 282 45.03725 291 125 724 17.26 287 129 285 45.26750 301 128 749 17.09 297 132 289 45.67775 311 130 774 16.79 307 134 293 45.73800 321 132 799 16.52 317 136 296 45.94825 331 135 824 16.38 327 139 302 46.03850 341 138 849 16.25 337 142 307 46.25875 351 140 874 16.02 347 144 310 46.45900 361 142 899 15.79 357 146 310 47.09925 371 145 924 15.69 367 149 312 47.76950 381 148 949 15.59 376 153 315 48.57975 391 150 974 15.40 386 155 316 49.051000 401 152 990 15.21 396 157 318 49.37
34
Chapter 3 : Server Consolidation using VM Migration
Table 3.5: Efficiency of NodMig and performance degradation of VMs at different valueof UTC and mAllowed
UTC fmAllow NUN NRN TVMG MT TET Eiff pDegVM0.6 0.25 6 0 2 6 55 0 6.50.6 0.50 6 0 2 6 55 0 6.50.6 0.67 5 1 4 12 61 25 11.80.6 0.75 5 1 7 21 70 14.3 180.6 0.90 5 1 10 30 79 10 22.80.65 0.25 6 0 2 6 55 0 7.10.65 0.50 6 0 2 6 55 0 7.10.65 0.67 5 1 5 15 67 20 14.550.65 0.75 5 1 5 15 67 20 14.550.65 0.90 4 2 11 33 83 18.18 25.840.7 0.25 6 0 3 9 57 0 11.050.7 0.50 5 1 4 12 61 25 13.770.7 0.67 4 2 6 18 68 33.33 18.530.7 0.75 4 2 6 18 68 33.33 18.530.7 0.90 4 2 10 30 79 20 26.580.75 0.25 6 0 3 9 57 0 11.840.75 0.50 5 1 4 12 61 25 14.750.75 0.67 4 2 7 21 70 28.57 22.50.75 0.75 4 2 7 21 70 28.57 22.50.75 0.90 4 2 11 33 83 18.18 29.820.8 0.25 5 1 3 9 57 33.33 12.630.8 0.50 4 2 7 21 70 28.57 240.8 0.67 4 2 7 21 70 28.57 240.8 0.75 4 2 7 21 70 28.57 240.8 0.90 4 2 11 33 83 18.18 31.80.85 0.25 5 1 4 12 61 25 16.720.85 0.50 4 2 5 15 67 40 19.030.85 0.67 4 2 5 15 67 40 19.030.85 0.75 3 3 6 18 68 50 22.50.85 0.90 3 3 11 33 83 27.27 33.80.9 0.25 5 1 4 12 61 25 17.70.9 0.50 4 2 5 15 67 40 20.150.9 0.67 4 2 5 15 67 40 20.150.9 0.75 3 3 6 18 68 50 23.820.9 0.90 3 3 12 36 88 25 36.820.95 0.25 5 1 4 12 61 25 18.680.95 0.50 4 2 5 15 67 40 21.270.95 0.67 3 3 5 15 67 40 21.270.95 0.75 3 3 6 18 68 50 25.140.95 0.90 3 3 12 36 88 25 38.86
3.4.2 Results
Results of NodMig and FFD algorithms are shown in Table 3.4. in terms of No of Used Nodes,
No of Released Nodes, No of Migrations, and Efficiency of both algorithms with different number
35
Chapter 3 : Server Consolidation using VM Migration
Figure 3.5: Graph between No of VMs and No of Released Nodes
of VMs at the interval of 25 VMs. Initially, we have some initial state of cluster. After server
consolidation, we generally find the different state of cluster from the initial state with the output
as defined in the objectives. State of the cluster can be viewed as shown in Table 3.4. after the
server consolidation by FFD and NodMig algorithms.
Number of migrations is equal to the total number of VMs mapping from a source PS to a
different destination PS during server consolidation. Finally, the efficiency of FFD and NodMig
algorithms are calculated using equation (3.15).
Fig. 3.4 shows the comparison graph between different number of nodes used by FFD and
NodMig algorithms at different number of VMs. As the number of VMs increases i.e. data center
becomes more bigger or no of simultaneous requests increases, the rate of increase in the number
of used PSs by NodMig algorithm decreases, while FFD shows the same rate of increase in the
number of used PSs. Simulation shows that at every increase in 200 VMs, the rate of increase in
no of used nodes decreases. But there is a limit in the decrease of rate of increase in the no of used
PSs. This will be done in the future work.
Fig. 3.5 shows the comparison graph between different number of released nodes resulted
by FFD and NodMig algorithms at different number of VMs. At any number of VMs, the num-
ber of released node by FFD(NRNffd) is less than or equal to the number of released node by
NodMig(NRNnmg).
36
Chapter 3 : Server Consolidation using VM Migration
Figure 3.6: Graph between No of VMs and No of Migrations
Figure 3.7: Graph between No of VMs and Efficiency
37
Chapter 3 : Server Consolidation using VM Migration
NRNffd ≤ NRNnmg (3.20)
For NodMig, As the number of VMs increases or decreases the rate of releasing node may
increase or decrease. It depends on the initial state of the cluster. Because of FFD is a heuristic
approach, the same case arises for FFD. But, generally rate of releasing node decreases as the data
center becomes more bigger.
Fig. 3.6 shows the comparison graph between different number of VM migrations resulted by
FFD and NodMig at different number of VMs. Here, we can observe the big difference between
the efficiency of FFD and NodMig in terms of VM migrations. Here, simulation shows that a
very less number of increase in the number of VMs as for 25,40,50,60, a small percentage in the
rate of migration efficiency of NodMig increases or decreases. But, for larger amount of increase
in the number of VMs, the rate of migration efficiency generally increases. In FFD, the rate of
migration efficiency either decreases or remains same for either large increase or small increase in
the number of VMs. This is the main advantage of NodMig over FFD.
Fig. 3.7 shows the comparison graph between overall efficiency of FFD and NodMig at dif-
ferent number of VMs. The overall efficiency of FFD and NodMig is calculated by Eqn. 3.5.
Simulation shows that the efficiency of NodMig always increases and the efficiency of FFD al-
ways decreases as the number of VMs increases. There is a threshold value for the increase in
the efficiency of NodMig. But, the rate of efficiency of NodMig may increase or decrease and the
same for FFD. The NodMig always shows the better efficiency than FFD.
3.5 Summary
Virtualization enables server consolidation to reduce the number of active PSs and to maximize
the utilization of pool of resources within the cloud data center. This results in the reduction of
cost for power, management, and cooling. VM live migration enables dynamic placement of VMs
across cluster of PMs.
Problem is formulated as a bin-packing problem where PMs replace bins and VMs replace
objects. Here, NodMig algorithm is proposed for server consolidation using VM live migration.
NodMig has three objectives: minimize number of active physical servers, minimize number of
VM migration, and maximize resources within the cloud data center. NodMig takes properties
from FFD, BFD, FF, and BF. But it differs from FFD that it has initial state of cluster. First, we
simulated NodMig with 15 VMs and 6 PMs and obtained optimal value of UTC and mAllowed.
Second, we simulated NodMig and FFD at the increasing rate of 25 VMs. We found NodMig is 2
to 3 times better than FFD, because NodMig has 2 more objectives than FFD.
38
CHAPTER 4
CONCLUSION AND FUTURE WORK
Energy efficient cloud data center is built to optimize each and every component of the cloud
data center in terms of its energy consumption. components like CPU, memory, hard disk, server,
network element, and cooling system of data center are considered for optimization. In this thesis
only CPU, memory, and server models and their optimization functions are created. Model and
optimization functions of Cooling system, network elements, and hard disk will be created in the
future work. By optimizing energy consumption of each component, cost that is incurred by the
data center is significantly reduced. This results in the reduction of carbon emission by the data
center and provides Green IT environment. This enables long life of the data center, at a time more
requests are serviced, and the level of QoS parameters increases.
Initially, VM placement and online VM migration technique that is based on heuristic ap-
proach FF, BF, FFD, and BFD is proposed. Problem statement is formulated as a knapsack prob-
lem. Optimization functions for processor, memory and server are separately defined. We are not
aware of the nature of upcoming request. For processing of request, VMs are run and at the same
time VMs are migrated according to the VM migration policy in heterogeneous environment. At
processing of each new request and some defined time interval, the overall energy consumption
and resource utilization of the data center are calculated. A number of VM placement techniques
exist. But, in this study VM placement and its migration are proposed to minimize energy con-
sumption, maximize resource utilization, and minimize number of VM migrations, and to reach a
certain limit of QoS parameters. We observed that proposed scheme shows the better result than
the brute force approach in terms of energy consumption, resource utilization, number of requests
processed by the data center. This result is dependent on the nature of requests within some time
interval and the configuration of data center.
Further, server consolidation is performed using VM migration. We are aware of the nature
of upcoming requests. We have some initial state of the cluster. VMs are migrated according to
the CPU utilization of server upto the threshold value, and maximum capacity of each resource
of server. VM migration stops when no further release of node is possible. VM migration is an
expensive operation. VM migration is only scheduled instead of actual VM migration at each
39
Chapter 4 : Conclusion and Future Work
iteration, because VM migration is an expensive operation. Actual VM migration takes place
if VM migration schedule releases at least one node. Another thing is that for a cluster of server
some predefined no of VM migrations should be take place for server consolidation. These 2 things
minimize the number of VM migrations with minimization of number of used nodes. NodMig is
compared with FFD. Fig. 3.4, 3.5, 3.6, and 3.7 show that NodMig always show better result than
FFD in terms of no of used nodes, no of released nodes, no of migrations, migration efficiency and
overall efficiency.
So, Energy consumption by the cloud data center is dependent on the type of the request,
at what time, in which situations or conditions, why, request is send, and configuration and the
quantity, the quality of resources of data center, what virtualization technique is being used by
the cloud data center, how big our cloud data center apart from algorithm proposed for server
consolidation, and model proposed for server, processor, memory, hard disk, network elements,
and cooling system.
First of all, We will further optimize the proposed model in comparison to the existing model,
and my proposed model. We will model the same or different problem based on GA, or SIP, or
PSO, or ILP. We will add another objective real-time scheduling, scalability, time-interval between
requests, migration time for VMs, network traffic within the data center, servicing time of request,
quickly changing the workload. Data will be taken as poisson distribution or t-distribution. We
would make private or hybrid cloud by considering these constraint with VMs placement and their
migration.
40
Bibliography
[1] Norman Bobroff, Andrzej Kochut, and Kirk Beaty. Dynamic placement of virtual ma-
chines for managing sla violations. In Integrated Network Management, 2007. IM’07. 10th
IFIP/IEEE International Symposium on, pages 119–128. IEEE, 2007.
[2] Alberto Caprara and Paolo Toth. Lower bounds and algorithms for the 2-dimensional vector
packing problem. Discrete Applied Mathematics, 111(3):231–262, 2001.
[3] Manogna Chebiyyam, Rashi Malviya, Sumit Kumar Bose, and Srikanth Sundarrajan. Server
consolidation: Leveraging the benefits of virtualization. Infosys Research, SETLabs Brief-
ings, 7(1):65–75, 2009.
[4] Jian-Jia Chen, Heng-Ruey Hsu, Kai-Hsiang Chuang, Chia-Lin Yang, Ai-Chun Pang, and Tei-
Wei Kuo. Multiprocessor energy-efficient scheduling with task migration considerations. In
Real-Time Systems, 2004. ECRTS 2004. Proceedings. 16th Euromicro Conference on, pages
101–108. IEEE, 2004.
[5] Jian-Jia Chen, Heng-Ruey Hsu, and Tei-Wei Kuo. Leakage-aware energy-efficient schedul-
ing of real-time tasks in multiprocessor systems. In Real-Time and Embedded Technology
and Applications Symposium, 2006. Proceedings of the 12th IEEE, pages 408–417. IEEE,
2006.
[6] Hui Cheng and Steve Goddard. Online energy-aware i/o device scheduling for hard real-
time systems. In Proceedings of the conference on Design, automation and test in Europe:
Proceedings, pages 1055–1060. European Design and Automation Association, 2006.
[7] Christopher Clark, Keir Fraser, Steven Hand, Jacob Gorm Hansen, Eric Jul, Christian
Limpach, Ian Pratt, and Andrew Warfield. Live migration of virtual machines. In Proceed-
ings of the 2nd conference on Symposium on Networked Systems Design & Implementation-
Volume 2, pages 273–286. USENIX Association, 2005.
[8] EN Mootaz Elnozahy, Michael Kistler, and Ramakrishnan Rajamony. Energy-efficient server
clusters. In Power-Aware Computer Systems, pages 179–197. Springer, 2003.
41
Bibliography
[9] Tiago C Ferreto, Marco AS Netto, Rodrigo N Calheiros, and Cesar AF De Rose. Server con-
solidation with migration control for virtualized data centers. Future Generation Computer
Systems, 27(8):1027–1034, 2011.
[10] Daniel Gmach, Jerry Rolia, Ludmila Cherkasova, and Alfons Kemper. Resource pool man-
agement: Reactive versus proactive or let’s be friends. Computer Networks, 53(17):2905–
2922, 2009.
[11] Taliver Heath, Bruno Diniz, Enrique V Carrera, Wagner Meira Jr, and Ricardo Bianchini.
Energy conservation in heterogeneous server clusters. In Proceedings of the tenth ACM
SIGPLAN symposium on Principles and practice of parallel programming, pages 186–195.
ACM, 2005.
[12] Chris Hyser, Bret Mckee, Rob Gardner, and Brian J Watson. Autonomic virtual machine
placement in the data center. 2008.
[13] Ravindra Jejurikar, Cristiano Pereira, and Rajesh Gupta. Leakage aware dynamic voltage
scaling for real-time embedded systems. In Proceedings of the 41st annual Design Automa-
tion Conference, pages 275–280. ACM, 2004.
[14] Si-Yuan Jing, Shahzad Ali, Kun She, and Yi Zhong. State-of-the-art research study for green
cloud computing. The Journal of Supercomputing, 65(1):445–468, 2013.
[15] Kyong Hoon Kim, Anton Beloglazov, and Rajkumar Buyya. Power-aware provisioning of
cloud resources for real-time services. In Proceedings of the 7th International Workshop on
Middleware for Grids, Clouds and e-Science, page 1. ACM, 2009.
[16] Dara Kusic, Jeffrey O Kephart, James E Hanson, Nagarajan Kandasamy, and Guofei Jiang.
Power and performance management of virtualized computing environments via lookahead
control. Cluster computing, 12(1):1–15, 2009.
[17] Young Choon Lee and Albert Y Zomaya. Energy efficient utilization of resources in cloud
computing systems. The Journal of Supercomputing, 60(2):268–280, 2012.
[18] David Meisner, Brian T Gold, and Thomas F Wenisch. Powernap: eliminating server idle
power. ACM SIGARCH Computer Architecture News, 37(1):205–216, 2009.
[19] Eslam Mohammadi, Mohammadbager Karimi, and Saeed Rasouli Heikalabad. A novel vir-
tual machine placement in cloud computing. Australian Journal of Basic and Applied Sci-
ences, 5(10):1549–1555, 2011.
42
Bibliography
[20] Aziz Murtazaev and Sangyoon Oh. Sercon: Server consolidation algorithm using live mi-
gration of virtual machines for green computing. IETE Technical Review, 28(3):212–231,
2011.
[21] M. Hemalatha N Ajith Singh. and. Basip: A virtual machine placement technique to reduce
consumption in cloud data center. Journal of Theoretical and Applied Information Technol-
ogy, 59(2):426–435, 2014.
[22] Eduardo Pinheiro, Ricardo Bianchini, Enrique V Carrera, and Taliver Heath. Load balancing
and unbalancing for power and performance in cluster-based systems. In Workshop on com-
pilers and operating systems for low power, volume 180, pages 182–195. Barcelona, Spain,
2001.
[23] Meikang Qiu, Laurence T Yang, Zili Shao, and Edwin H-M Sha. Dynamic and leakage
energy minimization with soft real-time loop scheduling and voltage assignment. Very Large
Scale Integration (VLSI) Systems, IEEE Transactions on, 18(3):501–504, 2010.
[24] Ivan Rodero, Juan Jaramillo, Andres Quiroz, Manish Parashar, Francesc Guim, and Stephen
Poole. Energy-efficient application-aware online provisioning for virtualized clouds and data
centers. In Green Computing Conference, 2010 International, pages 31–45. IEEE, 2010.
[25] Jyothi Sekhar, Getzi Jeba, and S Durga. A survey on energy efficient server consolidation
through vm live migration. International Journal of Advances in Engineering & Technology,
5(1), 2012.
[26] Ying Song, Yanwei Zhang, Yuzhong Sun, and Weisong Shi. Utility analysis for internet-
oriented server consolidation in vm-based data centers. In Cluster Computing and Work-
shops, 2009. CLUSTER’09. IEEE International Conference on, pages 1–10. IEEE, 2009.
[27] Frits CR Spieksma. A branch-and-bound algorithm for the two-dimensional vector packing
problem. Computers & operations research, 21(1):19–25, 1994.
[28] Jan Stoess, Christian Lang, and Frank Bellosa. Energy management for hypervisor-based
virtual machines. In USENIX Annual Technical Conference, pages 1–14, 2007.
[29] Akshat Verma, Puneet Ahuja, and Anindya Neogi. pmapper: power and migration cost aware
application placement in virtualized systems. In Middleware 2008, pages 243–264. Springer,
2008.
[30] Gregor Von Laszewski, Lizhe Wang, Andrew J Younge, and Xi He. Power-aware scheduling
of virtual machines in dvfs-enabled clusters. In Cluster Computing and Workshops, 2009.
CLUSTER’09. IEEE International Conference on, pages 1–10. IEEE, 2009.
43
Bibliography
[31] Frances Yao, Alan Demers, and Scott Shenker. A scheduling model for reduced cpu energy.
In Foundations of Computer Science, 1995. Proceedings., 36th Annual Symposium on, pages
374–382. IEEE, 1995.
[32] Andrew J Younge, Gregor Von Laszewski, Lizhe Wang, Sonia Lopez-Alarcon, and Warren
Carithers. Efficient resource management for cloud computing environments. In Green
Computing Conference, pages 357–364, 2010.
[33] Dakai Zhu, Rami Melhem, and Bruce R Childers. Scheduling with dynamic voltage/speed
adjustment using slack reclamation in multiprocessor real-time systems. Parallel and Dis-
tributed Systems, IEEE Transactions on, 14(7):686–700, 2003.
44